ABSTRACT
This paper is an empirical study on Harry Markowitz work on Modern Portfolio Theory. The
model introduced by him assumes the normality of assets return. We examined the OMX
Large Cap List1 by mathematical and statistical methods for normality of assets returns. We
studied the effect of the parameters, Skewness and Kurtosis for different time series data. We
tried to figure it out which data series is better to construct a portfolio and how these extra
parameters can make us better informed in our investments.
We have chosen 42 stocks from this list from different sectors of length 10 years. The complete Large Cap list
is available at appendix X.
Table of Contents
INTRODUCTION................................................................................................................................................. 1
DATA AND METHODOLOGY ................................................................................................................................ 2
THEORY ON MODERN PORTFOLIO THEORY .......................................................................................... 3
MODERN PORTFOLIO THEORY (MPT)................................................................................................................. 3
RISK AND REWARD (MEAN AND VARIANCE ANALYSIS) ..................................................................................... 4
A Short Note on Mean Calculation................................................................................................................ 5
Arithmetic Mean ............................................................................................................................................ 5
Geometric Mean ............................................................................................................................................ 6
Geometric Versus Arithmetic Mean............................................................................................................... 6
When to use Geometric Mean........................................................................................................................ 7
Variance and Standard Deviation ................................................................................................................. 7
Description of Standard Deviation in Portfolio theory ................................................................................. 8
Annualizing Returns and Standard deviation ................................................................................................ 8
MATHEMATICS OF THE MARKOWITZ MODEL ...................................................................................................... 9
DIVERSIFICATION .............................................................................................................................................. 14
Diversification in Markowitz model ............................................................................................................ 16
The Risk Free Asset ..................................................................................................................................... 18
THE SECURITY MARKET LINE (SML) ............................................................................................................ 18
THE CAPITAL MARKET LINE (CML) .............................................................................................................. 19
THE SECURITY CHARACTERISTIC LINE (SCL)................................................................................................ 19
THE CAPITAL ASSET PRICING MODEL (CAPM) ............................................................................................. 20
THE EFFICIENT FRONTIER AND MARKET PORTFOLIO ........................................................................................ 21
THE SHARPE RATIO........................................................................................................................................... 22
The Sharpe ratio in Portfolio theory ........................................................................................................... 23
SKEWNESS ........................................................................................................................................................ 23
KURTOSIS.......................................................................................................................................................... 24
CONSTRUCTION OF THE MODEL ON EXCEL......................................................................................... 25
EXCEL MODULES FOR PORTFOLIO MODELLING .................................................................................................. 25
PORTFOLIO RISK AND RETURN ......................................................................................................................... 25
USING SOLVER TO OPTIMIZE EFFICIENT POINTS................................................................................................. 26
FURTHER EXCEL IMPLEMENTATIONS ................................................................................................................. 27
Implementing the Portfolio VaR in Excel .................................................................................................... 28
EMPIRICAL INVESTIGATION...................................................................................................................... 30
THE JARQUE BERA TEST OF NORMALITY .......................................................................................................... 30
The Result of Jarque-Bera Test on Our Portfolio Assets............................................................................. 31
USING PLOTS TO MOTIVATE THE NON-NORMALITY OF ASSETS RETURN DATA .............................................. 32
Normal Probability Plot for Determining Non-Normality .......................................................................... 33
The Result of Normal Probability Plot on OMX Large Cap List................................................................. 35
THE PROBLEM WITH SHARPE RATIO AND THE REASON..................................................................................... 36
Adjustments to the Risk Regarding Higher Moments Effects ..................................................................... 38
CONSTRUCTION OF THE PORTFOLIO WITH NEW ADJUSTMENTS TO SHARPE RATIO ........................................... 39
ANALYSIS OF THE EMPIRICAL INVESTIGATION.................................................................................................. 39
Analysis for the different type of time series for constructing a portfolio ................................................... 44
SUMMARY OF THE RESULTS TOUCHED BY EMPIRICAL INVESTIGATION ............................................................ 50
CONCLUSION ................................................................................................................................................... 54
LIST OF REFERENCES ................................................................................................................................... 55
APPENDICES ..................................................................................................................................................... 57
APPENDIX 1 PROOF OF EXPECTED VALUE (MEAN) ........................................................................................ 57
APPENDIX 2 PROOF OF VARIANCE .................................................................................................................. 58
APPENDIX 3 TABLE OF SKEWNESS ................................................................................................................. 59
APPENDIX 4 TABLE OF KURTOSIS................................................................................................................... 60
II
III
Acknowledgement
We thank our supervisor Lars Pettersson, Asset Manager at IF Metall, for his support and
comments on our study.
IV
Introduction
The aim of this paper is to construct an empirical study on the concept, Modern Portfolio
Theory. The method is appreciated by scientists and without any doubt it is the most practical
investment model ever introduced. We are about to introduce the model and its components
fully, also covering the mathematical development of the model. But we are not going to give
just an introduction to this model.
The first part of this paper, Theory of the Modern Portfolio Theory gives a broad view on
the theory to the reader. Almost all the parameters and components of the basic model defined
in this part. We tried to be careful with references and choose the best literature in order to
give this opportunity to the reader to deepen his/her knowledge by referring to these sources.
Some historical facts, the risk and reward analysis, mathematical development of the model,
diversification and finally some other concepts introduced fully in this section of the paper.
The second part of this paper under title Construction of the Model on Excel shows how we
established the model on excel for further investigations. This part is brief and references
introduced can help the reader to get a better understanding of the process while referring to
the excel file provided by this study can also help the reader for these calculations.
The last section under title Empirical investigation is the main pat of this research. In the
first part we question the validity of one of the critical assumptions of the model and by some
statistical test we support our claim, then we introduce a new ratio to handle this inefficiency
regarding the model and finally we test these two ratios against each other by different
combination of some extra parameters introduced during the process.
In the following part, Data and Methodology we introduce the type of the data under use for
this study and some practical information about the data.
Consequently if all investors are seeking to maximize the utility, so all of them must behave
in essentially the same way! Which this consistency in behaviour can suggest a very specific
statement about their aggregate behaviour. It helps us to reach some description for future
actions. We will talk more about this in next sections.
Every model or theory is based on some assumption, basically some simplification tools.
Markowitz model relies on the following assumptions3;
By having these assumptions in mind, we will go through some concepts and terminologies
that will make us understand the model constructed in further part of this paper.
The assumptions are cited from the WebCab Components home page, the PDF file is available at internet
reference [2].
4
Internet Reference: http://facweb.furman.edu/~dstanford/mecon/b1.htm
A further assumption is that risk and return preferences of an investor can be described via a
quadratic utility function. This means when plotted on a graph, your utility function is a curve
with decreasing slope, for larger risk. Where w is an indicator for wealth and U is a quadratic
utility function. We have,
U w w w 2
A consumer's utility is hard to measure. However, we can determine it indirectly with
consumer behaviour theories, which assume that consumers will strive to maximize their
utility. Utility is a concept that was introduced by Daniel Bernoulli. He believed that for the
usual person, utility increased with wealth but at a decreasing rate. Figure 2 shows the utility
curve for investors with different risk preferences.
Risk aversion can be determined through defining the risk premium, which by Markowitz
defined to be the maximum amount that an individual is prepared to give up to avoid
uncertainty. It is calculated as the difference between the utility of the expected wealth and
the expected utility of the wealth.
U [ E ( w)] E[U ( w)]
This allows us to determine the characteristic of the behaviour of the investor regarding risk;
If U [ E ( w)] E[U ( w)] , then the utility function is concave and the individual is risk
averse;
If U [ E ( w)] E[U ( w)] , then the utility function is linear and the individual is risk
neutral;
If U [ E ( w)] E[U ( w)] , then the utility function is convex and the individual is risk
seeking.
It is what was defined by Markowitz (1959) and cited by Amenc et al [3]. Figure 2 gives a
graphical interpretation of what was stated above.
A Short Note on Mean Calculation
Before we move to the main challenge of MPT, the risk, we determine a method to calculate
the first parameter in use for constructing the model. It is possible to calculate mean of an
investment with several methods, but mainly arithmetic and geometric. We have chosen
geometric method and in following sections we motivate our choice by mathematical proofs
and examples. Before all these, we introduce them briefly;
Arithmetic Mean
The arithmetic mean of a list of numbers is the sum of all the members of the list divided by
the number of the items in the list.
1 n
1
a ai (a1 an )
n i 1
n
Where,
5
a Arithmetic mean
ai Sample's data where (i 1, 2, , n)
n Number of data set's memeber
Geometric Mean
The Geometric Mean of a collection of positive data is defined as the nth root of the product
of all members of the data set, where n is the number of members. The Geometric Mean of
the data set a1 , a2 ,..., an is:
1/ n
n
ai
i 1
Where,
n a1 a2 an
Equality holds if and only if, for every line a+bX that is tangent to g(x) at x=EX,
P ( g ( X ) a bX ) 1 .
This theorem can be used to prove the inequality between these two methods of averaging. If
a1 , a2 ,, an are positive numbers, defined as;
1
a A (a1 a2 an ),
(Arithmetic mean)
n
aG [a1a2 an ]1 n .
(Geometric mean)
Where an inequality relating these means is
aG a A .
In order to apply the Jensons Inequality, let X be a random variable with range a1 , a 2 , , an
and P ( X ai ) 1 , i 1, , n. since log x is a concave function, Jensens Inequality shows
n
that E (log X ) log( EX ); so,
log aG
1 n
1 n
log ai E (log X ) log( EX ) log ai log aA ,
n i 1
n i 1
So aG a A . 5
1 n
s
ai a
n 1 i 1
2
Where,
s The variance
n Number of the sample's members
ai The corresponding member of the data set where i 1, 2, , n
a The mean of the sample
s s2
The population standard deviation is denoted by 2 . The proof of the standard deviation
is similar to the variance, but it is squared.
For both the variance and standard deviations in these cases, they are assumed to be unbiased
estimators for , meaning that the random variables a1 , a2 ,....., an are assumed to be normally
distributed.
R annual m * R Periodic
annual m * Periodic
Where,
R annual Annualized return
m The number of periods per year
R Periodic The Periodic return
R annual 1 R Periodic
2
2
annual Periodic
1 R Periodic 1 R Periodic
2m
Where,
R annual Annualized return
m The number of periods per year
R Periodic The Periodic return
Figure 3 - Combinations of the risk less asset in a risky portfolio (Gruber et al. [9])
As we mentioned above, the ray discussed has the greatest slope. It can help us to determine
the ray. The slope is simply the return on the portfolio, RP minus risk-free rate divided by
standard deviation of the portfolio, P . Our task is to determine the portfolio with greatest
ratio of excess return to standard deviation . In mathematical terms we should maximize
(Later so called Sharpe ratio).
RP RF
P
X
i 1
Where X i s are the samples members, also can be random variables. The constraint can be
expressed in another way, Lintnerian, which considers an alternative definition for short sales.
It assumes that when a stock sold short, cash did not received but held as collateral. The
constraint with Lintner definition of short sales is6,
N
X
i 1
RF 1RF X i RF ( X i RF )
i 1
i 1
By stating the expected return and standard deviation of the expected return in the general
form we get,
N
X (R R
i 1
N N
X 2 2
X i X j ij
i
i
i 1
i 1 j 1
j i
Now we have the problem constructed and ready to solve. It is a maximization problem and
solved by getting the derivatives of the function with respect to different variables and
equating them to zero. It gives us a system of simultaneous equations,
10
1.
d
0
dX 1
2.
d
0
dX 2
N.
d
0
dX N
Where is called a Lagrange multiplier. Now we show how it proceeds and then its
application on our case;
1. Form the vector equation, f ( x) g ( x) .
2. Solve the system,
f ( x) g ( x)
g ( x) c
For x and . By extension of this problem we have n+1 equation in n+1
unknown x1 , x2 , x3 , , xn , ,
f x1 ( x1 , x2 , , xn ) g x1 ( x1 , x2 , , xn )
f x2 ( x1 , x2 , , xn ) g x2 ( x1 , x2 , , xn )
f ( x , x ,, x ) g ( x , x ,, x )
n
xn
1
2
n
xn 1 2
g ( x1 , x2 , , xn ) c
Where the solution for x ( x1 , x2 , , xn ) , along with any other point satisfying g ( x) c
and g ( x) 0 , are candidates for extrema for the problem.
11
N 2 2 N N
X i ( Ri RF ) X i i X i X j ij
i 1 j 1
i 1
i 1
j i
As it is written above, the ratio consists of multiplication of two functions. To derivate this
ratio we need to use Product Rule and as the second term suggests where it has power 1 2
another rule of calculus, the Chain Rule must be applied. After applying the chain rule, we use
product rule and we get,
3
N
2
N
N
N
d
1
X i ( Ri RF ) X i2 i2 X i X j ij 2 X k k2 2 X j kj
dX k i 1
i 1 j 1
i 1
2 i 1
j i
j k
N N
2 2
X i i X i X j ij
i 1
i 1 j 1
j i
Rk RF 0
N N
X 2 2
X i X j ij
i
i
i 1 j 1
i 1
j i
X i ( Ri RF )
i 1
N
X k k X j kj Rk RF 0
N N
i 1
X i2 i2 X i X j ij
jk
i 1
i 1 j 1
j i
12
X (R R
i 1
i 1
i 1 j 1
j i
X i2 i2 X i X j ij
Yields,
N
2
X k k X j kj Rk RF 0
i 1
By multiplication,
N
2
X k k X j kj Rk RF 0
i 1
Now by extension
d
( X 1 1i X 2 2 i X i i2 X N 1 N 1i X N Ni ) Ri RF 0
dX i
We use a mathematical trick, where we define a new variable Z k X k . The X k are fraction to
invest in each security, and Z k are proportional to this fraction. In order to simplify we
substitute Z k for X k and move variance covariance terms to the left,
Ri RF Z1 1i Z 2 2i Z i i2 Z N 1 N 1i Z N Ni
The solution of the above statement involves solving the following system of simultaneous
equations,
R1 RF Z1 12 Z 2 12 Z3 13 Z N 1 N
R2 RF Z1 12 Z 2 22 Z3 23 Z N 2 N
R3 RF Z1 13 Z 2 23 Z3 32 Z N 3 N
RN RF Z1 1N Z 2 2 N Z 3 3 N Z N N2
Now we have N equations with N unknowns. By solving for Zs we can get X k , which are the
optimum proportions to invest in stock k,
13
Xk
Zk
Z
i 1
Up to here we calculate the weights for the general form, where short sales are allowed and
lending and borrowing is possible. For other form of portfolio constructions, we follow the
same pattern but there might be other kinds of constraints defined.
Diversification
Despite what kind of role we have in finance world all of us might have heard this old English
proverb Do not put all your eggs in one basket by the character Sancho Panza in Miguel de
Cervantes Don Quixote8. It is simply what we call it here diversification. More specifically,
diversification is a risk management technique that mixes a wide variety of investments
within a portfolio. It is done to minimize the impact of any security on the overall portfolio
performance. A great reason for anybody to choose mutual funds is because they are said to
be well diversified. In order to have a diversified portfolio it is important that the assets
chosen to be included in a portfolio do not have a perfect correlation, or a correlation
coefficient of one.
Diversification reduces the risk on a portfolio, but not necessarily the return, and though it is
referred as the only free lunch in finance. Diversification can be loosely measured by some
statistical measurement, intra-portfolio correlation. It has a range from negative one to one
and measures the degree to which the various asset in a portfolio can be expected to perform
in a similar fashion or not.
Portfolio balance can be measured by some of these intra-portfolio correlations. As the sum
approaches negative one the percentage of diversifiable risk eliminated reaches 100%. It is
why it is called weighted average intra portfolio correlation. It is computed as9
Q
n
i 1
n
j 1
n
i 1
X i X j ij
j 1
Xi X j
Where,
Q is the intra-portfolio correaltion
X i is the fraction invested in asset i.
X j is the fraction invested in asset j.
http://www.riskythinking.com/articles/article13.php#herbison
Internet Reference: http://www.freepatentsonline.com/20030088492.html
14
Intra-portfolio correlation
1
0,75
0,5
0,25
0
-0,25
-0,5
-0,75
-1
Now lets come back again to diversification. In order to understand how to diversify a
portfolio we should understand the risk. According to Ibbotson et al. [11] risk has two
components, systematic and unsystematic. Where market forces affect all assets
simultaneously in some systematic manner it generates Systematic risk or what so called,
undiversifiable risk. Examples are Bull markets, Bear markets, wars, changes in the level of
inflation. The other component of risk is unsystematic one, or so called diversifiable risk.
These are idiosyncratic events that are statistically independent from the more widespread
forces that generate undiversifiable risk. The examples of a diversifiable risk are Acts of God
(Hurricane or flood), inventions, management errors, lawsuits and good or bad news affecting
one firm.
As defined above, Total risk of a portfolio is the result of summation of systematic and
unsystematic risks. On average, the total risk of a diversified portfolio tends to diminish as
more randomly selected common stocks are added to the portfolio. But, when more than
about three dozen random stocks are combined, it is impossible to reduce a randomly selected
portfolios risk below the level of undiversifiable risk that exists in the market. Figure 4
shows the graphical interpretation of this. The straight line separates the systematic risk from
unsystematic one, the systematic or undiversifiable risk lies under the straight line.
10
These figures from Table 1 is taken from; M. Statman, "How Many Stocks Make a Diversified Portfolio?"
Journal of Financial and Quantitative Analysis 22 (September 1987), pp. 353-64. They were derived from E. J.
Elton and M. J. Gruber, "Risk Reduction and Portfolio Size: An Analytic Solution," Journal of Business 50
(October 1977), pp. 415-37. Taken from Ross, Westerfield, and Jordan, "Fundamentals of Corporate Finance"
7th Edition (2006-11-14), pp. 406.
15
Figure 4 - The effect of number of securities on risk of the portfolio in the United States (Gruber et al [9])
x
i 1
E ( R p ) xi E ( Ri ) .
i 1
Where E ( Ri ) is the security analysts forecast for expected rate of return from the ith asset.
3. The Objective: Investment weights chosen by portfolio managers should add up to an
efficient portfolio which is:
16
The set of all efficient portfolios is called efficient frontier. This is the maximum return at
each level of risk. The efficient frontier dominates all other investment opportunities.
4. Portfolio Risk: In contrast with expected return of a portfolio which is based on forecast,
the risk of a portfolio is calculated from historical data available to the asset manager. The
risk of the portfolio, or its variance should be broken into two parts, the variance which
represents the individual risks and interaction between N candidate assets. This equation
(double summation) represents the variance-covariance matrix and can be expanded and
written in matrix form.
N
VAR( R p ) xi x j ij
i 1 j 1
Figure 5 - Relationship between expected return and standard deviation of return for various correlation
coefficients (Gruber et al. [9].)
17
Markowitz model gives the opportunity to the asset manager to combine a risky asset or a set
of risky assets (a portfolio of risky assets) with a riskless asset. In next parts this concept will
be more clarified when we explain all concepts in MPT one by one.
The Risk Free Asset
This asset is said to be a hypothetical asset which pays a risk-free return to the investor, with a
variance and standard deviation equal to zero. Usually this type of assets issued by the
government and can be referred to as government bond or Treasury bill (T-bill). But then it is
also assumed that government dose not go bankrupt. In reality we can also conclude that there
is no such thing as a risk-free asset, all financial instruments carry some degree of risk. But
also that these risk free-rates are subject to inflation risk. The common notation of the riskfree asset is RF .
SML
Here in the graph we can see that as the expected return increases so dose the risk (Beta). The
SML line is based on the risk free rate RF . We can then also see that since RF is risk free it has
a zero beta. When you go to the right of the graph, you will come to the market portfolio (M).
The market portfolio is a hypothetical portfolio, consisting of all the securities that are
available for an investor. That is why we have a beta of 1. The markets risk premium is
determined by the slope of the SML line.
18
E RM RF
The CML line also represents the highest possible Sharpe ratio possible. The CML line is
derived by drawing a tangent line from the intercept point of the efficient frontier (or the
optimal portfolio) to the point where the expected return equals the risk-free rate RF .
19
ri ,t i i RM ,t ei ,t
where,
i iM2
M
As we can see the beta is equal to the covariance between the return on asset i and the return
on the market portfolio iM , which is divided by the variance of the market portfolio M2 . This
also means that the risk-free rate has a beta of zero, the market portfolio a beta of one. We can
then define the CAPM model as follows:
E Ri RF i E RM RF
where,
E(R i ) = Expected return on asset i
R F = Risk-Free rate
i = Risk of asset i
E(R M ) = Expected return on the market
From the CAPM model, we can also establish that at equilibrium the return on asset, less the
risk-free rate; have a link to the return and the market portfolio which is linear. Also to note is
that the market portfolio is built according to the Markowitz principles. The graphical
representation of the CAPM is the security market line.
20
Optimal portfolios
should lie on this
curve (known as
the "Efficient
Fontier")
Return % (Mean)
In the figure we can se that the efficient frontier will be convex. The explanation is that there
is a risk and return characteristics of the portfolio that will change in a non-linear fashion as
its component weighting are changed. The case is also that the portfolio risk is a function of
correlation of the components assets, which also changes in a non-linear fashion as the
weighting of the component assets change.
The next step is finding the optimal market portfolio by connect some chosen risk-free asset
to the frontier, and then applying the Sharpe ratio which should be maximized. These two
properties will give you two points on the graph, which you then make a straight line from.
This line represents the lending part of a possible investment on the left side of the market
portfolio. If you draw the line straight to the right also, you will be able to borrow and invest
more in the market portfolio. This line that is connected to the efficient frontier is called the
capital allocation line (CAL).
21
The following figure shows a graphical view of what I just described. Here the E R stands
for return, is the standard deviation, RF is the risk-free asset and M stands for the market
portfolio.
where,
E RP RF
RP
RP
E RM RF
RM
This relation indicates that at equilibrium, this means that the Sharpe ratio of the portfolio to
be evaluated and the Sharpe ratio of the market portfolio are equal. The Sharpe ratio
11
William F. Sharpe won the Nobel Prize in economics for his development of the CAPM in 1990.
22
corresponds to the slope of the market line. If the portfolio is well diversified, then its Sharpe
ratio is close to that of the market.
The Sharpe ratio in Portfolio theory
The Sharpe ratio provides a good basis for comparing portfolios, and is widely used by
investment firms for measuring portfolio performance. In isolation, it dose not mean much.
This even when managers speak of good and bad Sharpe ratios, they are speaking only in
relative terms. E.g. if portfolio manager A has the highest Sharpe ratio of several managers,
then he or she has the highest risk-adjusted return of the managers for that period.
Skewness
Skewness is a parameter that describes asymmetry in a random variables probability
distribution. In other words a distribution is skewed if one of its tails is longer than the other.
Skewness can be positive; this means that it has a long tail in the positive direction. It also can
have a negative value, where it is called a negative Skewness. Consider the figure below
where two distributions are plotted by the same mean , and standard deviation , but the
one to the left is positively skewed (skewed to the right) and the one on the right is negatively
skewed (skewed to the left).
Skewness is equal to zero where we have a perfect asymmetry. Mathematically the kth
standardized moment, k , is defined by Finch [12] as,
2k / 2
12
3/ 2
2
3/ 2
23
Kurtosis
The fourth standardized moment according to the general form presented in the last section is
the Kurtosis defined by Finch [12] mathematically as,
4 4
22
Or equivalently it becomes,
4 4
22
3.
13
24
Portfolio return:
Matrix notation
Excel formula14
T
w e
=SUMPRODUCT(w, e)
Portfolio variance:
wTVw
Portfolio sigma
=MMULT(TRANSPOSE(w),MMULT(V, w))
wTVw
NOTE: that when computing the following models, the user needs to
press ctrl+shift+Enter for it to be executed.
For calculating the portfolios risk and return, we also need to compute some other parameters
that these are based on. These are the arithmetic mean, geometric mean, variance population,
standard deviation of population and the variance-covariance matrix. These are implemented
mainly by using excel user-defined functions already implemented in excel.
Arithmetic mean
Mathematical notation
1 n
xi
n i 1
Excel formula
=AVERAGE(arrays)
1/ n
Geometric mean
14
n
ai
i 1
=(GEOMEAN(arrays))-1
25
Variance of population
1 n
xi x
n i 1
=VARP(numbers)
1 n
( xi x)2
n i 1
Sigma of population
Covariance
1 n
xi x
n i 1
=STDEVP(numbers)
y y
i
=COVAR(array1;array2)
26
27
EFFICIENT FRONTTEAR
3,50%
3,00%
CAL = 0,2482x + 0,0033
RETURN
2,50%
2,00%
1,50%
1,00%
0,50%
0,00%
-1,00%
1,00%
3,00%
5,00%
7,00%
9,00%
11,00%
13,00%
15,00%
RISK
The essential inputs that are needed in this model are under the label Set up for the Modified
Sharpe. Then the first thing needed is the portfolio variance which is 0.000122351, next the
annualized portfolio standard deviation (Risk) 22.15%. Then after this we look up how many
standard deviations are needed to calculate VaR at the 95% level. This is obtained from the
normal distribution tables which will give you 1.64485, or in excel you use the built in
function NORMSINV (probability). This returns the inverse of the standard normal
cumulative distribution, which has a mean of zero and a standard deviation of one. Thereafter,
we multiply this by the portfolios standard deviation of 22.15%. This then means that we are
28
95% confident that our losses will not exceed 36.43% of our portfolio value. Below is the
spread sheet formula behind VaR calculation that was shown in figure 15.
Set up for Modefied Sharpe
Confidence level
No, Of standard deviations
Annual Standars deviation of Portfolio
Value at Risk of Portfolio
Value input
0,95
=NORMSINV($C$69)
=D64
=C70*C71
29
Empirical Investigation
This part under title Empirical Investigation tries to answer to some questions and use some
statistical methods to motivate these answers. In the last part of this paper we are going to
study some parameters on a group of constructed portfolios with up to forty two assets by
Markowitz model. Up to today lots of the financial models including Markowitz model were
subject to a series of assumptions. The pioneering work of Harry Markowitz in Modern
Portfolio Theory was not an exception; neither the semi-variance introduction could minimize
the damage of these assumptions. He defined the reward as expected return and the risk as the
standard deviation or variance of the expected returns. Rachev [15] claims since Markowitz
assumes the returns are normally distributed, the expectation operator is linear and the
portfolios expected return is simply given by the weighted sum of the individual assets
expected return. The variance operator, however, is not linear. This means that the risk of a
portfolio, as measured by the variance, is not equal to the weighted sum of risk of the
individual assets.
Before any further steps in analyzing the data we will examine the distributions normality of
our stream of data. There exist different statistical methods to do such a test. Some of them
are computational and it is easier to construct a Null Hypothesis Testing with the help of
them, and some others can only confirm our claim by visual evidence. We will here examine
the stream of data in two ways, Jarque-Bera test and QQ-plot.
Another interesting result of constructing a portfolio with Markowitz model was the
amazingly unrealistic results for the Sharpe ratio maximization. The problem with Sharpe
ratio is that it is accentuated by investments that do not have a normal distribution of returns.
As it is clear here, for a risk manager that tries to guard against large losses, the deviation
from the normality can not be neglected.
"In the case of testing the hypothesis that a sample has been drawn
from a normally distributed population, it seems likely that for large samples
and when only small departures from normality are in question, the most
efficient criteria will be based on the moment coefficients of the sample,
e. g. on the values of 1 and 2 .15"
E. S. Pearson, 1935
n 2 n
( 3) 2
6
24
Where,
3
1 n x x
i
The sample skewness
n i 1
4
1 n x x
i
The sample kurtosis
n i 1
1 n
xi x The sample variance
n 1 i 1
The result shows that under the hypothesis that xi is independent observations from a normal
distribution, for large n the distribution of the JB-test statistic is asymptotically Chi-square
distributed. This will help us to do a test on normality. If we have a large sample, and we
calculate the JB-test statistic on it and compare it with the null hypothesis that the data
represents a normal distribution, while we know that in 95% of the cases the value of the JBtest will be smaller than 5,99 for the normally distributed samples. Consequently we reject the
hypothesis of normality if the value of JB-test statistic exceeds this amount.
The Result of Jarque-Bera Test on Our Portfolio Assets
In order to see if we can reject the normality of the data set, we performed a JB-test on the
data sets. As mentioned before, our study compares 5 different sorts of data on OMX stock
exchange, daily, weekly, monthly, quarterly and yearly. The data provides a long term
investment of 10 years and what we did was to separate it into two 5-year period data for all
categories of data. In order to perform comparisons and analyze the results, we treated the
first period as the historical data and the second as the future one. Lets consider these
categories closer;
Daily returns are the longest set of the data we analyzed. The size of the data seems to have a
big impact on the JB-test. This claim becomes more touchable when we compare it more in
depth with other categories of data. As it is shown in the table taken from our empirical
investigation shown in Appendix 6, where the marked cells means that the null hypothesis of
normality is rejected we see that the statistic values for the JB-test are notably higher than
other categories in comparison with the daily returns. But more and less the number of stocks
that their normality can be rejected by this test is equal in the first three categories of data sets,
daily, weekly and monthly for both periods, historical and future one.
Surprisingly the quarterly data set has a larger number of normally distributed assets, which
can be due to the lack of data (the length of the data set is shorter than the latter categories).
31
But still it does not mean that we can apply models with normality assumptions on these data
sets, since almost 50-60 percent of the assets included in this category are not normally
distributed. The last category is the yearly data set, most of the assets successfully pass the
JB-test, but it can not be a reliable result considering the number of data in each data set. We
considered 10 years data, for two periods which will result in an analysis of a data set of five.
How convincing can the result of such a study be?! So we exempt this category from our
normality test by JB-test method.
Although JB-test rejects the normality for the data categories just mentioned with a good level
of significance, but in order to present a more decent result we decided to hire another method
for testing the normality. There is a visual method so called Q-plotting for normality, where
the data sets normality will be tested by plotting the data set against a normal distributed one.
What we did on our data was a large scale empirical investigation where we plotted the
histograms for forty four stocks on the Large Cap list of Stockholm Stock Exchange for two
5-year periods where data was sorted in 5 groups, daily, weekly, monthly, quarterly and
yearly. It gave us interesting results; the shape of the histograms did not support what we
expected; to have a nice bell shaped normal distribution. Instead we got all other possible
shapes. It was the reason why we started to calculate the 3rd and 4th statistical moments, the
parameters which are essential in forming the shape of the distributions. Here we present
some results, but the whole data analysis is available in the excel file provided by this report.
These histograms visualize distribution of data for four assets included in our empirical
investigation from the historical period (1997-2002).
32
Dis tribution
Dis tribution
300
250
250
200
200
Frequency
Frequency
150
150
100
100
50
50
%
%
%
%
%
%
%
5% 16% 78% 39%
71
33
94
00
39
78
16
,5
,
,
,
5,
3,
0,
1,
3,
5,
8,
-8
-6
-3
-1
-1 -50-1
-1
%
%
%
%
%
%
%
%
%
3% 22%
1
2
69
28
87
46
05
19
60
,6
,
,0
,4
9,
5,
0,
6,
2,
1,
5,
-7
-3
10
14
-2 -50 -2
-2
-1
-1
Return
Return
H&M
SAS
Distribution
Distribution
250
400
350
200
300
150
Frequency
Frequency
250
200
150
100
100
50
50
0
0
%
%
%
%
%
%
%
%
0% 48% 27%
8
9
33
12
91
94
16
37
-50
,7
,
,
,5
,7
7,
4,
0,
1,
5,
8,
-7
-4
-1
11
14
-1
-1
-1
%
%
%
%
%
9%
7% 85% 33% 81%
9%
23
75
27
78
30
,8
,3
,
,
,
,2
0,
1,
3,
4,
6,
-8
-7
-5
-4
-2
-1
-50
Return
Latour B
Return
Lundberg
As the figures show clearly, some of the tails are showing fat-tail and no strong visual sign for
asymmetric distributions. Some of the stocks like Latour B show an excess Kurtosis and
heavy tails while others like SAS shows an unsymmetrical distribution. But still we need a
more convincing method to become sure whether asset returns under our study are normally
distributed or not?
Another visual method used widely by statisticians is normal probability plotting. This can be
a good method to visualize the data distribution to study. We consider that the reader of this
paper is already familiar with normal plotting. We just mention different cases that we might
face in our data plot. Some pictures will be provided in order to help reader to understand the
concept better.
Normal Probability Plot for Determining Non-Normality
It is a technique to see whether the data is approximately normally distributed or not. The
normal probability plot is constructed on a graph with two axes, where vertical axes are our
data and the horizontal is the z-values. With the help of normal probability plot we can answer
these two questions;
16
Reference: Taken from the empirical investigations done under this study.
33
We can conclude that the normal probability plot shows a strong linear pattern, where
the minor deviations from the line are insignificant. The normal distribution is a good
model fitting the subject data.
2- Data have fat-tails;
If we face a plot with such a shape we conclude that, the normal probability plot shows a
non-linear pattern. Consequently the normal distribution is not a good model for these data.
If we take a closer look, both fat and short tail distributed data share some common
characteristics. Both showing a S-shaped curve in the middle and both are deviated form the
reference line at both ends of the plot.
17
18
34
In this case the normal probability plot shows a reasonably linear pattern in the middle of
the plot but it departures from both ends. As we suggested for long fat-tailed one, another
distribution other than normal would be reasonable for this case.
4- Data are skewed right;
Here we have a strong non-linear pattern, where normal distribution is not a good pattern for
this data set.
The Result of Normal Probability Plot on OMX Large Cap List
After introducing the possible cases, we plotted our data on OMX Large Cap list, for 10 years
period where data is categorized in 4 different groups, daily, weekly, monthly and quarterly. It
was not far from our expectation that almost all the plots confirmed the nonlinearity and the
fact that these data are not normally distributed. Here we present the normal probability plots
of the stocks we showed histogram plots for earlier. We can take a look at the shapes and
analyse them in relation with the histograms just mentioned.
19
20
35
0,2
0,2
0,1
0,15
0,1
0
-4
-3
-2
-1
-0,1
0,05
SAS
H&M B
-4
-0,2
-3
-2
0
-1 -0,05 0
-0,3
-0,1
-0,15
-0,4
-0,2
Z-Score
Z-Score
H&M B
SAS
0,3
0,1
0,2
0
-4
-3
-2
-1
-0,1
Lundberg
L atour B
0,05
0,1
0
-4
-3
-2
-1
-0,05
-0,2
-0,3
Z-Score
Latour B
-0,1
Z-Score
Lundberg
Figure 21 - The normal probability plot for 10 years daily asset returns21.
The histograms showed a high-peaked data plot and made us suspicious about the asset
returns being normally distributed, and lead us to conclude that our data sets are fat-tailed.
Also based on the calculations for skewness and kurtosis which are the parameters of statistic
value of the JB-test, it is highly convincing to reason and claim the non-normality of the
distributions. Now the normal probability plot confirms strongly that the fundamental
assumption, normality of the asset returns by Markowitz model is not true. Most of the
investment managers are not aware of this fact that each one of this assumptions are made to
simplify the calculations, where the normality assumption by Markowitz model makes it easy
for almost every investment manager despite the level of knowledge in mathematics or
statistics to construct the model. The damage of this assumption often is underestimated by
investors or basically can be unknown to them.
Reference: Taken from the empirical investigations done under this study.
36
used in constructing the portfolio. For a portfolio with assets which are normally distributed
the volatility automatically infer the extreme loss risk. This illustrated under a paper work
released by Sinopia Asset Management member of the HSBC Group by Bertland [18]22. But,
is the asset returns always normally distributed? The answer is No. What we showed so far by
this empirical investigation by JB-test and graphs was that the assets returns are not normally
distributed, at least in the case of Large Cap List on OMX.
The following graph compares the distribution of monthly returns of a Fixed-Income
Arbitrage instrument and the equivalent normal distribution. Volatility is equal for both cases,
while the extreme losses on the Fixed-Income Arbitrage instrument are much higher than the
Gaussian one. This shows the danger of relying solely on volatility to measure risk in
portfolios.
Figure 22 - Extreme losses in normal VS. Non-normal asset distributions with equal volatility23
A feeling which is common to every risk manager is that the perception of investment is
naturally related to the potential for extreme losses. Despite the technical analysis and
complicated calculations, all investment managers are concerned about the large losses, what
we called here extreme losses. By the graph just presented, we are convinced that normal
distribution is not a good model to fit the assets returns. Lets consider another graph
presented on a study by Rachev et al [19] and see what happened to Dow Jones Industrial
Average (DJIA) daily return between 1991 and 2003. In this graph presented in Figure 23, the
volatility plotted in the interval just mentioned. As you see this graph indicates that extreme
losses happened where it will never occur under normal distributions assumptions.
Now lets see how these facts can affect Sharpe ratio. Because of the way volatility is defined,
returns that are 5% above average will inevitably be treated in the same way as those 5%
below average. When we consider a normal distribution since its skewness is zero and
perfectly symmetrical it can not cause any problem regarding the results. But where the
symmetrical distribution is not the case, it can be a source of problem. Consider the two
extreme losses occurred at left tail of the distribution shown in Figure 22. It shows the
22
23
37
extreme losses, now considers these extreme occurrences that are on the positive side of the
mean, and then the investment has extreme gains. Compared to normally distributed ones
presented in the same graph but still the risk indicated by the volatility is equal for all three
cases. Our conclusion would be that the volatility is not an adequate measure for risk. In the
same way accuracy of Sharpe ratio can be questioned!
(R p -R f )
MVaR
Where,
1
1
1
MVaR W [ {zc ( zc2 1) S ( zc3 3zc ) K (2 zc3 5 zc ) S 2 } ],
6
24
36
And,
24
38
R p Portfolio return
R f Risk-free ratio
MVaR Modified VaR
Standard Deviation
S Skewness
K Kurtosis
z c is the critical value of the probability (1- )
The benefit of involving the higher moments in our calculation is that we want to avoid
underestimating risk. MVaR is introduced here since the conventional value at risk exhibit the
same shortcoming as the standard deviation (based on normal distribution assumptions for
asset returns). For derivation of the formula you can refer to Favre and Galeano [2002]25.
25
26
39
40
becomes first while the Sharpe quote stands at second place after the portfolio with stocks
sorted out by positive skewness. When considering the future data risk, this category comes as
best as a result of diversification (access to more assets to be included in the portfolio).
Based on this study the worst performance is for the category which hires stocks with both
Kurtosis grater than 3 and positive Skewness. Comparing Sharpe ratios calculated by both
methods for both periods this category stand for the lowest results. This category is ranked
last.
Weekly Portfolio
In the same pattern we analyzed the data for the daily portfolios constructed, we can look at
weekly ones. We have the same table for Sharpe ratios, return and risk of portfolios when
different factors, skewness and kurtosis considered in our optimization.
Lets consider the first portfolio constructed by the traditional Sharpe ratio where skewness
and kurtosis effect were not considered by the original model. Not surprisingly the traditional
Sharpe ratio is almost double the Modified one in the first case, but by referring to the excel
file we will see that the stocks chosen and their weights are identical for both portfolios. This
case introduces the second highest return for the historical portfolio and the lowest risk. But it
will be interesting to compare the results with the future portfolios. In this case the Sharpe still
has the highest value and risk and return kept their positions.
The next portfolio is the one with both positive skewness and kurtosis greater than 3. In this
case we have the lowest Sharpe ratio for both historical and future portfolios. Returns are
second best, but considering the high risks they are not worth to consider. But it is interesting
to consider the velocity of losing value of the returns from the historical portfolio to the future
one, from 23,21% to 9,12%.
The third portfolio is the one with just positive skewness. It has still a traditional Sharpe ratio
greater than 1, highest return and simultaneously lowest risk. The most interesting case, but
lets see if these characteristics remain the same when we go to the future portfolio. Sharpe is
still relatively high. The return is not the highest but the risk managed to be the lowest for the
future data.
41
The fourth portfolio that we analyzed is the portfolio with stocks which have kurtosis greater
than 3. As it is predictable by looking back again in the second case it is not a good method to
construct a portfolio. Low Sharpe ratio for both periods, and the risk which is high for both
periods and the return which is not so high compared with other cases for the first period, but
interestingly not diminished as much as other portfolios for the second period.
Again as the daily analysis, the worst case seems to be the second case. But it is difficult to
distinguish between the first and third case as a candidate for the best case.
Monthly Portfolio
In the following data set, we can se that values for the two portfolios with Skewness and
Kurtosis and Kurtosis greater than 3 is not included. The reason for this is that the numbers
of stocks after sorting out for these portfolios were not reaching the desired level for an
efficient diversification. This is one of the most important concepts of portfolio construction.
Considering the first available category which is the portfolio constructed with the traditional
Markowitz model where only mean and variance are considered, the Sharpe ratio is the
highest for both methods of calculation, modified and traditional Sharpe. When we move to
future portfolios for the same category the Sharpe ratios almost became half. When we are
analyzing the Sharpe ratios, it would make more sense to look at risk and return closely.
Return is still highest for this category while representing the least risk. But surprisingly while
the return became almost half of the historical portfolios the risk is decreased only by 6%. The
next category is where we have stocks included in the portfolio with only positive skewness.
In this category we have almost the same figures as the last case, but in general 1-2 percent
less.
The interesting result is where we compare the modified Sharpe ratio by the traditional one
where the portfolio constructed by the traditional Markowitz model. In this category we have
a minimization of only 43 percent for modified Sharpe ratio against 52 percent of the
traditional case. In order to analyze this result, you can compare these figures with the case of
considering stocks with positive skewness. In the case of constructing a portfolio with only
positive skewness, the figures for both methods of calculation of the Sharpe ratios are
42
identical. It clarified that the modified Sharpe ratio considers the positive skewness even in
the case of traditional Markowitz model.
Quarterly Portfolio
Based on the same reason we mentioned on the last type of the portfolio, we have only two
categories to analyze for quarterly portfolio.
The first category is where we have the general model applied. Looking at Sharpe ratios and
their developments we will see that this category has the highest Sharpe ratios both traditional
and modified while the development for the traditional case is worse compared with other
categories, a figure equal to almost 80%.
Compared to the case of the portfolio with positive skewness, the return considering the risk
for the same category is not at all satisfying 14% of return versus 15% of risk. The
development of the risk in this category considering the development of its counterpart is not
good at all.
Considering the portfolio with stocks which has only positive skewness, we have a good
Sharpe ratio for both methods compared with the first category analyzed where the return is
almost 26% and risk is relatively low, only 16%. Development of the figures from the
historical portfolios to future is interesting. While return diminished, the risk has risen for
both categories.
The other phenomena explained above, about the development of modified Sharpe ratio for
the original method of the construction of the portfolio is touchable here too. The
development of the Modified Sharpe ratio remained almost constant for the case of the
portfolio with positive skewness while it has fallen for the first category.
43
Yearly Portfolio
For this category, since the time series in not long, we can not construct portfolios with
reasonable structures. So we wont consider this category in our investigation. The reason for
this is the unrealistic results of Sharpe ratio calculations.
Analysis for the different type of time series for constructing a portfolio
Daily time series:
We have constructed 4 different types of portfolios by combining two additional parameters,
skewness and kurtosis. The time series length is 10 years that is divided into two periods.
These two periods as mentioned before are called historical and future, each of length of 5
years. In this part we will compare the development of Sharpe ratio, risk and return from
historical period to future one for these 4 types of portfolios. The development parameterized
by ratios presented in Table 3.
The first portfolio is constructed by standard Markowitz model, and the first part of the table
shows the traditional Sharpe ratio development against modified Sharpe ratio. For the daily
portfolio we have a traditional Sharpe of 0.99 which is relatively high. The development is
about 54% decrease comparing with weekly portfolio, which has a traditional Sharpe ratio of
1.12 for the historical period and a negative development of 37%. This can be due to the level
of diversification, but surprisingly we see that the number of the stocks included in both
portfolios is the same. By looking at risk it can be explained why we have such a
development. Since the number of data is less in the case of weekly portfolio, and probably
most of the extreme events happened during the week and this data are not available at
weekly time series to analyze, we have a lower risk for weekly portfolio versus the daily one
and consequently a higher traditional Sharpe ratio for the weekly portfolio. To see if it is
really true we can compare the modified Sharpe ratio of the two portfolios. The historical
daily portfolio has a Sharpe ratio of 0.6 while the weekly has a modified Sharpe ratio of 0.68.
By looking at Value at Risk for both portfolios we can see that the daily portfolio has a higher
value compared with the weekly portfolio, 0.36 versus 0.3. This explains the differences in
the modified Sharpe ratios obtained by these two data sets and the reason is that the daily
44
portfolio as a matter of fact considers all the events in the time series, including positive
extreme events. So as a result we have a higher return for the daily portfolio compared with
weekly one, but also higher risk. It will not obviously violate the concept of the utility. This
result becomes more touchable if the reader understands fully what we explained under
subtitle The Problem with Sharpe Ratio and the Reason.
The second row of the table shows the development for Sharpe ratios on a portfolio
constructed considering two additional parameters, positive skewness and a kurtosis greater
than three. Since the number of stocks is limited under such a constraint the traditional Sharpe
ratio has a low value. It is due to the level of diversification for this category. The annual
portfolio risk is 26% for the daily portfolio which assigns weight to only 8 stocks to the
portfolios out of 19, while in the weekly case it is 37% which is the result of a portfolio of
only 2 stocks out of 11. Since the weekly portfolio has such a high risk and low Sharpe
consequently and the construction of the portfolio is impossible for other categories, a deeper
analysis of this type of portfolio (with positive skewness and kurtosis greater than 3) seems to
be useless. But among all other portfolios if we consider the development of Sharpe ratio, risk
and return, still this type of portfolio is the worst one, with the highest decrease of Sharpe
ratio and return for the two time periods versus the lowest decrease of the risk. So we can
conclude that this type of portfolio not only in Daily time series but also in other time series
can not be a considerable investment. So we will omit this category from our analysis for
further time series.
The third row representing a portfolio constructed with only positive skewness. The
development among the other portfolios can be claimed to be moderate. This category enjoys
a better change in Share ratio and particularly the return for the two periods compared with
the traditional Markowitz model. In the case of a daily time series, the traditional Sharpe ratio
is 1,04, and greater than one which is good. And still in a weekly time horizon the Sharpe
ratio have a value grater than 1. Considering the values of the return and risk, they show still
the best performance compared with other portfolios in this time horizon.
The fourth type of portfolio is the one with only kurtosis greater than three. Before looking
into the Sharpe ratio and the return and risk, we will start by considering the number of stocks
included in this type of portfolio to have a rough estimation of the level of diversification.
Among the stocks on the Large Cap list on OMX, there are 29 stocks that have kurtosis
greater than 3. After optimization for this category, the weights are assigned to only 9 of
them. This resulted in a high return but also high risk. Explanation for the high return is due to
the positive Kurtosis which results from fat-tail distributed asset returns and this can be seen
when we apply the new Sharpe ratio by a modified risk measure, resulting in a lower Sharpe
ratio, with the same risk, return and weights for the portfolio. But the question is that if this
category is a good type of investment. Answering this question is difficult, since we have a
relatively low negative development in Sharpe ratio compared with other portfolio types in
the same time horizon but also still a high level of risk. This can be considered as the last
choice to an investment manager since it has the lowest traditional or modified Sharpe ratio.
So in order to compare, we can not only take a look at development ratios, but also the
magnitude of these parameters are essential to our analysis.
The last interesting result which is true for all 4 different portfolios is the result of
optimization for maximizing both traditional and modified Sharpe ratio. Although we get the
same weights, portfolio risk and return for these cases, the modified Sharpe ratio is smaller
45
than the traditional. This confirms our claim about the overestimation of the Sharpe ratio
calculated by the traditional method.
Daily Portfolio - Sharpe
Increase/Decrease %
Optimization
Sharpe Ratio Modified Sharpe Ratio Sharpe Ratio Modified Sharpe Ratio Sharpe Ratio Modified Sharpe Ratio Sharpe Ratio Modified Sharpe Ratio
-54,39%
-61,48%
-52,78%
-48,35%
-54,39%
-61,43%
-52,75%
-48,34%
-37,46%
-51,84%
-43,66%
5,44%
-37,45%
-51,84%
-43,64%
5,51%
-52,17%
N/A
-52,50%
N/A
-43,92%
N/A
-52,50%
N/A
-79,62%
N/A
-55,67%
N/A
-53,04%
N/A
-55,67%
N/A
Sharpe Ratio Modified Sharpe Ratio Sharpe Ratio Modified Sharpe Ratio Sharpe Ratio Modified Sharpe Ratio Sharpe Ratio Modified Sharpe Ratio
-59,82%
-64,37%
-58,01%
-53,78%
-59,82%
-64,34%
-57,99%
-53,78%
-45,09%
-60,71%
-49,97%
-24,22%
-45,10%
-60,71%
-49,96%
-24,16%
-51,89%
N/A
-51,41%
N/A
-51,88%
N/A
-51,42%
N/A
-66,93%
N/A
-44,41%
N/A
-66,90%
N/A
-44,41%
N/A
Sharpe Ratio Modified Sharpe Ratio Sharpe Ratio Modified Sharpe Ratio Sharpe Ratio Modified Sharpe Ratio Sharpe Ratio Modified Sharpe Ratio
-22,92%
-21,95%
-20,53%
-18,55%
-22,92%
-21,96%
-20,53%
-18,55%
-15,37%
-29,37%
-16,24%
-25,88%
-15,39%
-29,37%
-16,24%
-25,86%
-6,03%
N/A
-4,50%
N/A
-6,02%
N/A
-4,50%
N/A
27,64%
N/A
21,44%
N/A
27,79%
N/A
21,45%
N/A
46
Daily Portfolio
Weekly Portfolio
Monthly Portfolio
Quarterly Portfolio
0,992020696
0,603105761
0,45248599
0,275087437
1,12922393
0,686519981
0,706233886
0,429416123
1,414414883
0,859903226
0,676540412
0,48223651
1,493233205
0,907821329
0,304372119
0,426347861
0,86085468
0,523362572
0,331603596
0,201847344
0,521497735
0,317048111
0,2511566
0,15269236
N/A
N/A
N/A
N/A
N/A
N/A
N/A
N/A
1,043982831
0,634696239
0,49297677
0,299879543
1,08602251
0,6602549
0,611900981
0,372109973
1,322041672
0,803744036
0,627952737
0,381771894
1,39841282
0,850174664
0,619887812
0,376864504
0,880049544
0,535032139
0,454549224
0,276372552
0,556742372
0,338475322
0,587054964
0,357137817
N/A
N/A
N/A
N/A
N/A
N/A
N/A
N/A
47
The second portfolio in this time category is constructed with stocks which have only positive
skewness. As it was not far from imagination, we have again the same level of decrease in the
both portfolios, optimized by traditional Sharpe ratio or by the modified Sharpe ratio. This
can be the result of our choice of stocks, the stocks with only positive skewness. It can be
seen that the Modified Value at Risk used to measure the risk for constructing the modified
Sharpe ratio can consider the right skewed effect, since still giving the same level of the risk
and return for both of the portfolios optimized by traditional or modified Sharpe ratio. You
can compare the results for Sharpe ratios by table 7.
Quarterly Time Series:
The last time series we are going to analyze more in detail is the quarterly time series. The
data to construct two types of portfolios was not available to us as indicated in the table by
N/A. This type of time horizon is quiet different from other time horizons. The reason is the
release of quarterly reports by companies. Almost all companies try to clean up their financial
losses and show a good performance, although it might come quiet late into the analysis of
investors, but it has its impact on stock markets, both on liquidity and volatility of the market.
As it is shown in table 6, the first row where we have a portfolio constructed by traditional
Markowitz model for quarterly time horizon, youll see that the decrease in traditional Share
ratio is the highest change ever in our research. The reason can be clarified by looking more
closely into the risk factor. For this time horizon the difference in return of the portfolios is
almost in the same range of the other time horizons, that is why we exempt this parameter and
go directly to the risk for finding out the reason for this dramatic decrease in traditional
Sharpe ratio.
The pattern of changes in the difference of ratios for risk which started from daily time
horizon just turned the sign and became an increase for quarterly time horizon. This increase
in the risk can be due to release of the quarterly reports by corporations and of course
followed by an increase in trade for stocks. This results in more liquidity in the market. The
other reason can be the cumulative return of the stocks during the quarter, while we ignore the
volatility of the market in this period. We should also consider positive or mostly
overestimated effect of these reports; the annual returns based on figures deviated long from
the mean, and the annual risk based on the not so frequent return statistics, but cumulated and
long away from the mean.
Now lets take a look at the second portfolio in this category. It takes only stocks with
positive skewness. For historical time horizon we have almost the same risk, return and
Sharpe ratio for these two portfolios. These figures are available in tables 7, 8 and 9 for
comparison. But the interesting results come into eyes when we look at developments of these
two portfolio types in future time horizon. The portfolio with positive skewness has a less
difference in risk development in compare with the Markowitz model and also a much less
difference in returns developments. This consequently follows by a less difference in
traditional Sharpe ratio. In contrast with the latter portfolio, the one with skewness shows the
same development for traditional vs. modified Sharpe ratio.
To conclude one more time we can claim that the portfolio constructed by stocks with positive
skewness can generate better results for future periods.
48
Historical
Future
Historical
Future
Historical
Future
Historical
Future
Risk
Traditional Markowitz Model
Sharpe Ratio
Modified Sharpe Ratio
Sharpe Ratio
Modified Sharpe Ratio
Positive Skewness and Kurtosi Greater than 3
Sharpe Ratio
Modified Sharpe Ratio
Sharpe Ratio
Modified Sharpe Ratio
Positive Skewness
Sharpe Ratio
Modified Sharpe Ratio
Sharpe Ratio
Modified Sharpe Ratio
Kurtosis Greater than 3
Sharpe Ratio
Modified Sharpe Ratio
Sharpe Ratio
Modified Sharpe Ratio
Daily Portfolio
Weekly Portfolio
Monthly Portfolio
Quarterly Portfolio
22,15%
22,15%
17,08%
17,08%
18,63%
18,64%
15,76%
15,77%
15,38%
15,38%
14,45%
14,45%
15,38%
15,38%
19,63%
19,65%
26,72%
26,71%
20,85%
20,85%
37,10%
37,10%
26,21%
26,21%
N/A
N/A
N/A
N/A
N/A
N/A
N/A
N/A
22,22%
22,22%
17,66%
17,66%
19,62%
19,62%
16,43%
16,43%
15,74%
15,74%
15,03%
15,03%
16,30%
16,30%
19,80%
19,80%
23,93%
23,93%
19,49%
19,49%
29,42%
29,42%
21,81%
21,81%
N/A
N/A
N/A
N/A
N/A
N/A
N/A
N/A
Historical
Future
Historical
Future
Historical
Future
Historical
Future
Return
Traditional Markowitz Model
Sharpe Ratio
Modified Sharpe Ratio
Sharpe Ratio
Modified Sharpe Ratio
Positive Skewness and Kurtosi Greater than 3
Sharpe Ratio
Modified Sharpe Ratio
Sharpe Ratio
Modified Sharpe Ratio
Positive Skewness
Sharpe Ratio
Modified Sharpe Ratio
Sharpe Ratio
Modified Sharpe Ratio
Kurtosis Greater than 3
Sharpe Ratio
Modified Sharpe Ratio
Sharpe Ratio
Modified Sharpe Ratio
Daily Portfolio
Weekly Portfolio
Monthly Portfolio
Quarterly Portfolio
25,81%
25,81%
10,37%
10,37%
24,90%
24,91%
13,67%
13,68%
25,62%
25,62%
12,33%
12,33%
25,78%
25,78%
8,52%
8,53%
26,83%
26,83%
9,56%
9,57%
23,21%
23,21%
9,12%
9,12%
N/A
N/A
N/A
N/A
N/A
N/A
N/A
N/A
27,03%
27,03%
11,35%
11,35%
25,17%
25,17%
12,59%
12,60%
24,67%
24,67%
11,99%
11,99%
26,66%
26,66%
14,82%
14,82%
24,89%
24,89%
11,50%
11,51%
20,24%
20,24%
15,34%
15,35%
N/A
N/A
N/A
N/A
N/A
N/A
N/A
N/A
49
50
But to distinguish which portfolio(s) are not clear investment objectives was not as difficult.
The least attractive investment would be the on optimised by assets which are both positively
skewed and have a kurtosis grater then 3.
Monthly Portfolio
For this data set, we did not include portfolios with Skewness and Kurtosis and Kurtosis
grater then 3. This is due strongly to the inefficiency of diversification possibility.
Then when we only head two portfolios to look at, traditional Markowitz and positively
skewed. There is not really a clear distinction between optimized Markowitz model or the
positively skewed one. But still again it leans a little more against the Markowitz model.
But also to notice an interesting finding in this part of the investigation is that the traditional
Sharpe ratio and the modified one differs, but only for the traditional Markowitz model and
not for the portfolio with positive skewness.
Quarterly Portfolio
In this part we omitted the two portfolios mentioned also in the monthly data. Since the
diversifiable amount of assets is not acceptable.
Then for the traditional Markowitz portfolio we see that it has the highest Sharpe ratio
compared to the positively skewed one. We also saw in this case that the traditional Sharpe
had an 80% in difference compared to the modified one of 53%. While for the skewed
portfolio was the same in both cases.
One could also notice from our investigation that the risk for both portfolios has risen when
we look in to the future. While at the same time the return has decreased considerably more
for the Markowitz model then the skewed portfolio. But looking at the historical figures only,
would make it difficult to determine which portfolio performed the best. But looking at the
whole picture of the future and history, we can say that the positively skewed one is the best
choose.
Yearly Portfolio
For this type of series we have conclude from the attempt to implement the different models,
that we are not able to construct reasonable portfolio structures with this data. The reason is
the unrealistic result of the Sharpe ratio calculations.
2. Compare the portfolios calculated in part one with other portfolios that
have different time series (monthly, daily, etc).
Daily time series
We can see in this data set that the standard Markowitz model has a Sharpe of 0.99 while
weekly has a Sharpe of 1.12, this is higher then the daily. But the number of stocks in the
portfolio is the same. We found that the explanation for this lies under the risk comparison.
Since the data set is larger in the daily series, more extreme events can be registered, then
51
during the weekly data. This is showed by the daily data having a higher risk and weekly a
lower one. We could also find the effect on the modified Sharpe, which had daily Sharpe of
0.60 and weekly 0.68, and the VaR was 36% and 30% respectively. This shows that the
difference is smaller since VaR accounts for extreme events.
In our time series investigation we have noticed that by building portfolios during a then year
period divided by two five years period, and adding on the different combinations of
skewness and kurtosis. That the smaller amount of time series you have to work with the less
diversification possibility you will have when adding on extra constraints to the Markowitz
model. This is also because it limits the amount of stocks to be included in the optimisation.
This is the reason why we omitted some portfolios, because they did not follow the
diversification principle.
The different portfolios have different levels of impact depending on the time series. Mostly
because the methods we used had effects on the assets at different time series events.
We were also able due to the time series to conclude that, for our 4 types of portfolios that
although the weight, return and risk, modified Sharpe ratio is smaller than the traditional one.
This shows that our claim about the overestimation of the traditional Sharpe ratio is true.
Weekly time series
In this time series we found that the Sharpe ratio for the Markowitz model had the smallest
change, as we can see in table 6 comparing all other portfolios and time series. The weekly
time series shows an overall smallest change in Sharpe compared to the other time series.
We were able to find one portfolio compared to other time series and portfolios, which had an
increase in the future, Sharpe ratios. The reason for this was that the decrease in risk in the
portfolio was greater then the decrease in the return. The portfolio in this case was the one
with a kurtosis greater then 3.
The positively skewed and the Markowitz portfolio is the two clear chooses in this data set.
But when comparing to the different time series, we could se that the Markowitz portfolio had
the lowest return compared to the other time series for this portfolio, but had the second
highest risk. The skewed portfolios had the second highest risk but the third highest return. In
the future data for these portfolios the Markowitz model had the third highest risk and the
highest return, compared to its other time series. For the skewed one we had the third highest
risk and the second highest return. So in this case it would lean a little more in favour for the
Markowitz model.
Monthly time series
In this data series we omitted the portfolios with positive skewness and kurtosis greater than
3, and the one with only kurtosis greater than 3. The reason for this was that it did not fill the
concept of diversification.
What we found here was that, for the traditional Markowitz model when looking ahead was
that the level of decrease in the traditional and modified Sharpe is not equal. But also the
modified Sharpe showed a more stable result. A reason for this could be the low level of risk
associated with that type of portfolio. The monthly portfolio has a lower level of risk
52
compared to the other time series daily and weekly, whereas the return was approximately the
same.
There was also an obvious pattern realised on the characteristics of the behaviour of risk. This
was that the bigger frequency of time series we had, the higher the level of risk, due to
discrepancies. In the other way around it could decrease the level of risk, which it did for the
monthly and quarterly data.
For the portfolio with positive skewness, we obtained again the same level of decrese in both
portfolios, optimizing using traditional and modified Sharpe ratio. It was shown that the
Modified Value at Risk used to measure the risk for constructing the modified Sharpe ratio
can consider the right skewed effect, since it still gives the same level of the risk and return
for both of the portfolios optimized by traditional or modified Sharpe ratio.
Quarterly time series
We realized that one of the underlying reasons for the large shifts in the stock returns is, from
how the stock markets are affected by the speculators and analysts that play the market during
quarterly reports.
We can find that in this time series the decrease in the traditional Sharpe is the highest. And
that it had the highest Sharpe ration in the historical time series than the other once. It is also
interesting to see in the investigation that the risk when looking into the future has a positive
change (increase), then all the other time sets. We believed that the markets speculations on
the quarterly reports had an impact of large effects on the stock prices. But that the increase in
the future risk, is affected of other things mentioned.
53
Conclusion
After introducing the model and constructing it on excel we started our empirical
investigation on the model. The results were amazingly satisfying in this section. The tests on
the normality of the assets return showed a strong support for our claim about the
abnormality of the assets return with a high level of statistical significance. The next part as a
result of the abnormality of data sets leads us into finding a new measure for the performance
in order to consider these inefficiencies. The new measure introduced and compared with the
original model also in combination with some new parameters, like skewness and kurtosis.
Hiring these parameters changed the results of the future portfolios and we presented above
which factor had the greatest influence or how they could make an investment manager better
in his investment.
Form our empirical research we are able to draw the following conclusion of the study we
made. The description will be brief since a deeper explanation of what we conclude in our
investigation could be found in the empirical and analysis part of the paper.
The data set which replicated best for the future portfolios was the monthly time
series. It showed a moderate accurate estimate of the future, when risk and return was
taken into account.
54
List of References
[1] Gallati, Reto, Risk Management and Capital Adequacy, Blacklick, OH, USA: McGraw-Hill
Companies, 2003.
[2] URL: http://www.webcabcomponents.com/Java/documentation/WebCab_Portfolio_J2SE.pdf, last
observed 12:57, 8 June 200727.
[3] Amenc Noel, and Veronique Le Sourd, Portfolio Theory and Performance Analysis, 2nd ed.,
Johnson Wiley & Sons Ltd, England, 2005.
[4] Casella George, and Berger Roger L., Statistical Inference, 2nd ed., Thomson Learning, 2001.
[5] Wackerly Dennis, Mendenhall Williams, Schaffer Richard L., Mathematical Statistics with
Applications, Duxbury Press; 6th edition, 2001.
[6] Markowitz Harry, Portfolio Selection: Efficient Diversification of Investments, 2nd ed., Blackwell
Publishing Limited, 1991.
[7] Chincarini Ludwig B., and Daehwan Kim, Quantitative Equity Portfolio Management An Active
Approach to Portfolio Construction and Management, McGraw Hill, 2006.
[8] M. Statman, How Many Stocks Make a Diversified Portfolio?, Journal of Financial and
Quantitative Analysis no. 22, September 1987.
[9] Elton Edwin J., Gruber Martin J., Brown Stephan J., and Goetzmann William N., Modern Portfolio
Theory and Investment Analysis, 6th ed., Wiley, 2005.
[10] Stewart James, Calculus, 5th ed., Brooks Cole, 2002.
[11] Francis Jack Clark, and Ibbotson Roger, Investments A Global Perspective, Pearson Higher
Education, 2002.
[12] Finch, S. J. Testing for Normality, New York, NY, USA: Marcel Dekker Incorporated, 2001.
[13] Jackson Mary, and Staunton Mike, Advanced modelling in finance using Excel and VBA, Wiley
Finance, 2001.
[14] Cormac Butler, Mastering Value at Risk, 1st ed., Prentice Hall, 1999.
[15] Rachev Svetlozar T., Handbook of Heavy Tailed Distributions in Finance, 1st ed., Elsevier,
Netherlands, 2003.
[16] Rachev Svetlozar T., Menn Christian, and Fabozzi Frank J., Fat-Tailed and Skewed Asset Return
Distributions, 1st ed., John Wiley & Sons, Inc., Hoboken, New jersey, 2005.
[17] Lhabitant, Franois-Serge. Hedge Funds: Quantitative Insights. Hoboken, NJ, USA: John Wiley
& Sons, Incorporated, 2004.
27
Note that all internet references referred in this paper holds up to this date. Change in the address of these
internet references is unavoidable.
55
[18] Bertrand Jean-Charles, and Berlemont Damien, In the land of hedge fund, the Sharpe ratio is no
longer king, Sinopia Asset Management, 2005.28
[19] Rachev Svetlozar T., and Martin Doug, Stable Non-Gaussian Risk Management and Factor
Models, FinAnalytica, Sofia, 2004.29
[20] Greg N. Gregoriou, and Jean-Pierre Gueyie, Risk-adjusted performance of funds of hedge funds
using a modified Sharpe ratio, The Journal of Wealth Management, Winter 2003.
28
29
56
Appendices
Appendix 1 Proof of Expected Value (Mean)30
Assume some population X and some random sample of that population x1 , x2 ,......, xn . Then
we can define another random variable Y , which is he mean of the sample:
Y
1 n
xi
n i 1
For population in statistics one defines the mean of X as . We can then prove that the mean
of Y is:
1 n
E Y E xi
n i 1
1 n
E xi
n i 1
1 n
E xi
n i 1
1 n
n i 1
1
n
n
This means then that since the expected value of Y is , this then is referred to as an
unbiased estimator of . One can also refer to it as, if we dont know , but we know the
sample mean, x , then we should use x to estimate .
30
57
n 1 i 1
n
2
1
E xi x
n 1 i 1
1 n
xi x
n 1 i 1
1 n
2
E xi 2 E xi x E x
n 1 i 1
1 n
1
1 n 2
2 E xi xi 2
n 1 i 1
n j 1
n
E x
n
j 1 k 1
xk
1 n 2 2 2 2
n n
n 1 i 1
2
n
n 1
1
n 1 i 1
n
n 1 2
n 1
This proves that s 2 is an unbiased estimator of the population variance. The proof for the
standard deviation is the same but instead the whole expression is squared E s 2 2 .
31
58
Benchmark
Benchmark
JM
Castellum
Kinnevik B
Swedbank A
Fabege
Latour B
Lundberg
OMX
Ratos B
resund
SEB A
SHB A
Investor B
Hufvudstaden A
Industryvrden A
SKF B
Volvo B
Skanska B
Trelleborg B
NCC B
Peab B
Hexagon B
Seco Tools B
Assa Abloy B
SAS
Securitas B
Atlas Copco A
Sandvik
Holmen B
SCA B
SSAB A
Electrolux B
H&MB
Tele2 B
Ericsson B
Nokia
Getinge B
Elekta B
Meda A
Axfood
Swedish Match
Vostok Nafta
SSVX 30
SIXRX (TR)
Finance
Finance
Finance
Finance
Finance
Finance
Finance
Finance
Finance
Finance
Finance
Finance
Finance
Finance
Finance
Industry
Industry
Industry
Industry
Industry
Industry
Industry
Industry
Industry
Industry
Industry
Industry
Industry
Material
Material
Material
Commodities
Commodities
Telecommunication
IT
IT
Health Care
Health Care
Health Care
Commodities
Commodities
Energy
Daily
0,781
0,053
0,071
0,465
0,187
0,150
1,084
0,063
-0,097
3,722
-0,008
-1,495
0,148
0,463
0,244
0,701
0,040
0,423
0,218
-1,962
0,137
-0,348
0,973
-0,318
-0,082
0,554
-0,331
0,965
0,415
0,438
-1,264
0,320
0,494
0,023
-0,860
0,443
-0,042
0,335
0,085
1,171
2,330
1,491
0,212
0,858
Historical
Future
Weekly Monthly Quarterly Yearly Daily
Weekly Monthly Quarterly Yearly
0,113 -0,447
-0,741 -1,369
1,957
0,922
0,848
0,904
0,966
0,411
0,086
0,313
1,292
0,102
-0,664 -0,523
-1,926 -2,018
0,601
0,409
0,483 -0,580
-0,330
-0,011 -0,524
-0,779 -0,605
0,504 -0,140
-0,603
0,281
-0,185
-0,345 -0,401
0,169 -0,340
0,178
0,615
0,749
0,577
0,107
0,225
1,726
0,364
1,156
0,308
0,632
1,116
0,294
-0,021
-0,459 -0,152
-0,737 -1,320
0,992
0,313
0,113 -1,050
-1,433
-0,990 -1,123
-0,767
1,466
0,794
0,888
1,025
0,511
1,085
-0,289
0,346
-1,344 -0,654
-0,451 -1,272
-1,010 -1,893
-0,487
-0,539
0,475
0,108
1,860
0,733
1,195
1,051
0,156
1,741
1,783
0,683
-0,384
0,385
0,875 -0,086
0,091 -0,889
0,133
-0,443 -0,332
-0,175
1,060
-1,423 -0,755
0,072
0,359
-0,318
0,622 -0,173
-0,256
0,351
0,749 -1,075
-1,233
0,434
0,527
0,035 -0,534
-0,342 -1,545
0,721 -0,087
0,313
0,788
0,008
-0,335 -0,084
-0,944 -1,405
0,073 -0,620
-0,963 -0,530
0,028
-0,391 -0,829
-2,601 -1,267
1,124
0,237
-0,597 -0,059
-0,181
-0,196 -0,264
-0,397 -0,028
0,068 -0,108
0,145
1,120
-0,076
-0,539 -0,297
-1,487 -2,011
0,296
0,214
-0,067
0,912
0,474
0,014 -0,147
0,216
1,619
0,708
0,212
0,282
0,314
0,007
-0,135
0,122
-1,186 -0,952
-1,171
0,117
-0,533
0,767
0,112
0,098
0,360
-1,150 -0,753
0,914
0,273
-0,420 -0,188
0,144
-0,314
0,035
-0,859
0,811
0,099 -0,068
-0,229
1,714
-0,375
-0,289 -0,031
0,236 -0,531
0,366
0,415
-0,218
1,438
-0,057
-0,044 -0,105
-0,339 -1,662
0,300 -0,092
0,088
0,252
0,331
0,273
0,032
-0,548
1,320
0,050 -0,485
-0,167 -0,184
-0,598
-0,471
0,007
0,737 -0,234
0,712
0,663
-0,054 -0,659
0,064
0,296
0,671
-0,733 -0,545
-0,950 -0,616
-1,021
0,682
0,450
0,393
0,276
-0,161
0,528
0,464
0,553
0,633
0,829
-1,079
-0,393 -0,639
-0,873 -0,715
0,306
0,407
0,242
1,062
0,506
-0,268 -0,049
-1,355 -0,460
0,093
0,412
-0,233
1,827
0,106
-0,357 -0,020
0,925 -0,982
-0,679 -0,900
-0,751
2,211
-0,823
-0,869 -0,922
0,090 -0,446
0,203
0,390
-0,729
0,252
0,381
-0,084 -0,060
-0,078
1,599
0,162 -0,134
0,109
0,900
0,209
0,166
0,005
-0,836
0,307
0,030
0,211
0,168 -0,589
0,389
-0,205
0,087
-0,743 -0,042
-0,007
0,500
0,088 -0,470
0,617
0,055 -0,383
-0,014 -0,832
0,071
0,520
1,163
0,437
0,267
0,264
1,792
1,514
0,870
-0,086 -0,078
0,784
1,422
0,356
0,715
2,412
0,115 -0,620
-0,260 -0,216
0,749
0,549
-0,342
0,023 -0,251
-0,405
0,434
0,642
0,286
0,140
1,266
0,407
0,275
0,306
0,308 -0,189
0,799
1,059
0,566
1,799
-0,051
0,016 -0,503
-0,853
0,434
3,342
3,424
1,485 -0,868
1,676
3,641
2,519
0,298
0,747
0,804
1,105
0,297
1,457
-0,345
0,485
0,123
-0,224 -0,479
0,635
0,653
0,411
0,807
0,146
-0,172
0,098
1,163
0,334
1,088 -0,095
1,189
0,487
0,213
0,443
0,541
1,767
1,249
32
These tables presented on the appendices are the results of our study and they are available on the excel file
provided by this report.
59
Benchmark
Benchmark
JM
Castellum
Kinnevik B
Swedbank A
Fabege
Latour B
Lundberg
OMX
Ratos B
resund
SEB A
SHB A
Investor B
Hufvudstaden A
Industryvrden A
SKF B
Volvo B
Skanska B
Trelleborg B
NCC B
Peab B
Hexagon B
Seco Tools B
Assa Abloy B
SAS
Securitas B
Atlas Copco A
Sandvik
Holmen B
SCA B
SSAB A
Electrolux B
H&MB
Tele2 B
Ericsson B
Nokia
Getinge B
Elekta B
Meda A
Axfood
Swedish Match
Vostok Nafta
Future Data
Daily
Weekly Monthly Quarterly Yearly
4,195 0,104 -0,179
-0,048 -0,285
3,949 2,668
1,657
4,861
4,180
4,621 2,833
0,385
0,244 -3,259
3,147 1,529
0,421
-0,237 -1,497
7,701 2,721
8,145
0,457
1,656
2,790 0,738
0,424
0,894
2,009
17,046 3,537
1,644
-0,467
2,326
31,342 1,366
0,682
2,349 -2,213
4,376 2,508
0,596
-0,621
3,440
19,433 12,302
3,090
1,255
1,474
3,830 1,995 -0,249
-0,241
1,819
6,289 6,457
1,177
0,124 -1,844
4,470 1,577
0,363
0,053
2,962
4,033 2,025 -0,266
0,512
2,311
3,782 2,602
1,292
8,319
2,653
4,749 3,544
0,917
-0,781 -2,160
3,627 0,682
1,355
3,175
4,314
5,219 0,398 -0,604
-0,329
2,578
2,828 0,511
1,211
3,465
1,168
3,689 1,398
1,048
1,358
1,637
3,097 1,054 -0,674
0,084 -1,373
4,557 1,045 -0,333
-0,472 -1,338
3,850 1,648 -0,299
0,559
2,652
2,339 0,121
0,457
0,258
2,115
14,096 2,000
1,750
0,391 -1,457
7,577 3,076
2,142
-0,186 -1,138
2,931 1,099
2,518
-0,716 -0,870
20,055 6,058
0,779
-0,058 -0,419
4,057 0,075
0,193
2,678
0,364
3,510 0,423
0,043
0,257 -0,369
12,904 3,017
0,901
0,154 -3,022
4,141 1,192
0,462
-0,729
3,003
5,471 3,130 -0,699
1,707 -2,063
5,648 1,668 -0,554
0,689 -2,629
7,233 1,568 -0,244
-0,113 -0,100
5,837 2,342
8,576
3,415
0,148
7,212 5,377 13,606
1,298
0,637
5,751 2,157
1,400
0,421
1,706
2,133 1,009
0,309
0,097 -0,291
5,043 2,633
1,548
2,443
0,168
14,137 29,876 12,279
0,245 -1,251
12,387 2,835
0,629
-0,680 -3,017
3,121 0,877 -0,635
2,072 -0,922
4,986 3,212
0,231
5,266
0,636
60
Benchmark
Benchmark
JM
Castellum
Kinnevik B
Swedbank A
Fabege
Latour B
Lundberg
OMX
Ratos B
resund
SEB A
SHB A
Investor B
Hufvudstaden A
Industryvrden A
SKF B
Volvo B
Skanska B
Trelleborg B
NCC B
Peab B
Hexagon B
Seco Tools B
Assa Abloy B
SAS
Securitas B
Atlas Copco A
Sandvik
Holmen B
SCA B
SSAB A
Electrolux B
H&MB
Tele2 B
Ericsson B
Nokia
Getinge B
Elekta B
Meda A
Axfood
Swedish Match
Vostok Nafta
Daily
SSVX 30
SIXRX (TR)
Finance
Finance
Finance
Finance
Finance
Finance
Finance
Finance
Finance
Finance
Finance
Finance
Finance
Finance
Finance
Industry
Industry
Industry
Industry
Industry
Industry
Industry
Industry
Industry
Industry
Industry
Industry
Industry
Material
Material
Material
Commodities
Commodities
Telecommunication
IT
IT
Health Care
Health Care
Health Care
Commodities
Commodities
Energy
0,261
0,166
0,957
0,604
0,326
0,333
0,294
1,019
0,420
0,292
0,789
0,497
0,828
0,222
0,960
0,640
0,619
0,522
0,458
0,352
0,305
0,439
0,316
0,721
0,312
0,591
0,689
0,596
0,553
0,475
0,597
0,675
0,901
1,240
2,029
1,629
0,370
0,552
0,274
0,297
0,179
0,677
Historical
Weekly Monthly Quarterly Yearly
0,313
0,130
1,482
0,516
0,370
0,632
0,314
1,217
0,686
0,284
0,966
0,470
0,792
0,271
1,038
0,794
0,775
0,640
0,564
0,570
0,219
0,478
0,397
0,872
0,483
0,669
0,797
0,620
0,553
0,477
0,596
0,731
0,907
1,626
2,013
1,379
0,263
0,556
0,207
0,322
0,099
0,779
0,109
0,129
1,618
0,472
0,217
0,708
0,310
1,877
0,517
0,473
0,943
0,280
0,704
0,295
1,072
0,674
0,521
0,515
0,416
0,505
0,122
0,388
0,265
0,924
0,534
0,714
0,645
0,624
0,618
0,382
0,566
0,701
0,808
1,658
2,246
1,367
0,220
0,438
0,182
0,045
-0,185
0,789
0,211
0,285
1,788
0,426
0,307
0,702
0,225
2,044
0,420
0,674
0,851
0,289
0,842
0,396
0,995
0,601
0,405
0,380
0,329
0,414
0,154
0,280
0,230
0,853
0,456
0,630
0,455
0,435
0,679
0,383
0,532
0,870
0,647
1,795
2,407
1,990
0,218
0,227
0,183
0,148
-0,193
0,832
0,432
-0,193
1,111
0,040
-0,082
0,196
-0,024
1,258
0,144
0,189
0,212
0,011
0,659
0,208
0,972
1,011
0,301
0,571
0,176
0,772
-0,408
-0,325
0,190
0,770
0,168
0,598
0,601
0,973
0,849
0,412
0,672
0,656
0,962
1,268
2,574
2,869
-0,435
-0,578
0,092
-0,595
-0,327
0,579
Daily
0,552
0,267
1,169
0,815
0,416
0,523
0,378
1,387
0,548
0,323
1,181
0,864
1,196
0,368
1,000
1,028
0,989
0,934
0,828
0,623
0,467
0,617
0,484
1,152
0,845
1,147
1,243
0,971
0,667
0,657
0,786
1,026
0,812
1,156
2,165
1,299
0,564
0,615
0,515
0,400
0,272
0,940
Future
Weekly Monthly Quarterly Yearly
0,668
0,344
1,211
0,815
0,517
0,855
0,422
1,782
0,702
0,499
1,068
0,796
1,378
0,477
1,075
1,036
1,037
1,076
0,935
0,753
0,573
0,827
0,491
1,329
1,100
1,247
1,322
1,050
0,688
0,666
0,930
1,012
0,736
0,958
2,210
1,169
0,670
0,651
0,630
0,554
0,205
0,743
0,797
0,656
1,869
0,755
0,411
0,902
0,419
2,037
0,710
0,493
0,755
0,641
1,327
0,673
1,126
0,808
0,996
1,184
1,056
0,671
0,687
1,070
0,575
1,225
1,190
1,231
1,383
0,581
0,770
0,536
0,710
0,924
0,529
1,038
2,971
1,070
0,802
0,873
0,683
0,419
-0,097
1,022
1,201
0,952
2,116
0,864
0,640
0,947
0,523
1,971
0,894
0,703
0,751
0,627
1,281
0,792
1,076
0,639
0,880
1,186
1,149
0,877
0,797
1,159
0,619
1,292
1,322
1,034
1,144
0,517
0,672
0,564
0,912
0,773
0,376
1,031
2,516
0,611
0,968
0,973
1,220
0,396
0,307
1,375
1,048
0,256
1,841
0,718
0,155
0,596
-0,010
1,771
0,558
0,406
0,846
0,594
1,400
0,311
0,983
0,291
0,880
0,836
0,721
0,902
0,005
0,884
0,421
0,587
1,008
0,716
1,184
0,660
0,365
0,049
0,790
0,568
0,552
0,899
2,023
0,837
0,549
0,773
1,220
-0,398
-0,067
1,770
61
Benchmark
Benchmark
JM
Castellum
Kinnevik B
Swedbank A
Fabege
Latour B
Lundberg
OMX
Ratos B
resund
SEB A
SHB A
Investor B
Hufvudstaden A
Industryvrden A
SKF B
Volvo B
Skanska B
Trelleborg B
NCC B
Peab B
Hexagon B
Seco Tools B
Assa Abloy B
SAS
Securitas B
Atlas Copco A
Sandvik
Holmen B
SCA B
SSAB A
Electrolux B
H&MB
Tele2 B
Ericsson B
Nokia
Getinge B
Elekta B
Meda A
Axfood
Swedish Match
Vostok Nafta
Historical Data
Future Data
Daily
Weekly Monthly Quarterly Yearly Daily
Weekly Monthly Quarterly Yearly
SSVX 30
245,26
12,95
27,06
8,62
1,85
830,74 115,46
29,21
9,42
3,03
17,12
7,60
23,12
2,89
1,68
46,75
18,34
6,52
13,72
3,68
SIXRX (TR)
Finance
282,49
44,71
17,65
6,40
2,31
151,99
0,28
17,86
7,52
8,47
418,62
15,19
23,68
8,95
1,02
7,86
25,84
16,41
7,94
4,31
Finance
292,84
11,51
12,69
6,53
5,48
1098,89
2,74
86,38
5,25
1,49
Finance
Finance
137,14
3,71
3,60
7,07
6,83
2,27
58,35
15,14
4,95
1,66
2172,07
65,99
21,51
9,05
0,97 10198,25
41,23
15,48
10,78
1,89
Finance
Finance
898,32
25,89
12,93
3,47
0,84 40096,46
29,41
13,17
5,73
6,02
Finance
44,65
8,09
20,13
12,89
3,05
141,04
13,73
15,03
9,87
2,92
101300,83
49,59
14,41
5,60
3,15 14002,58 971,71
4,22
2,73
0,61
Finance
Finance
424,87
316,78
12,48
7,90
2,00
37,73
17,59
24,74
7,97
1,23
12600,81
496,50
10,96
6,22
1,61
556,87 132,16
7,75
6,40
4,99
Finance
55,27
151,50
11,13
11,08
4,89
162,33
19,86
18,21
6,87
1,99
Finance
Finance
98,99
26,13
19,76
4,94
2,85
52,97
13,69
24,06
7,32
1,74
16,05
40,65
13,93
24,00
2,43
30,47
7,54
12,74
41,51
1,36
Finance
Finance
369,19
72,43
18,79
11,79
5,49
158,36
4,39
10,39
11,20
5,55
59,99
39,65
24,29
0,09
1,37
20,67
64,01
6,89
6,66
3,73
Finance
69,26
46,42
28,13
8,33
2,30
288,95
66,32
29,42
8,45
2,22
Industry
Industry
37,11
23,01
32,80
0,40
6,59
1,48
61,39
7,33
4,38
1,45
14129,62
162,39
12,55
2,88
6,33
26,05
25,50
9,74
5,99
0,86
Industry
Industry
145,06
38,07
6,88
6,91
4,26
4,58
40,92
30,37
8,59
4,53
Industry
38,43
26,62
26,98
9,20
2,47
148,18
40,69
25,01
9,21
4,15
612,13
75,53
28,02
4,61
2,02
36,53
17,98
24,58
4,81
2,33
Industry
Industry
122,81
7,04
9,17
5,66
7,26
43,42
84,06
14,56
6,54
1,61
3201,82
42,85
25,87
5,19
0,47
6180,56
18,47
3,51
6,73
4,18
Industry
60,26
21,72
14,48
7,62
7,46
1040,48
3,49
5,71
9,22
3,81
Industry
Industry
93,42
222,11
7,62
13,48
5,35
40,44
41,42
1,21
10,44
3,35
495,54
11,66
7,21
8,22
0,67 14665,29
97,62
14,78
9,30
2,86
Industry
Industry
102,81
69,46
6,92
0,25
3,60
106,17
86,61
17,76
5,58
1,62
45,65
65,85
9,68
5,81
2,90
15,15
70,02
19,68
8,21
3,17
Industry
9401,69
90,36
8,92
7,77
4,84
5002,06
29,58
17,57
6,10
7,72
Material
Material
119,72
104,01
20,40
12,02
6,01
93,46
32,27
14,53
10,45
2,13
49,73
34,00
20,47
1,29
2,04
311,58
1,24
30,79
3,35
5,42
Material
Commodities
73,98
28,03
24,25
4,09
3,34
378,04
19,01
28,48
5,66
6,60
Commodities
7226,67
4,11
6,41
7,29
6,35
964,70
20,20
24,99
7,27
2,58
228,03
10,13
14,98
4,19
7,75
413,63
6,98
98,87
7,00
2,33
Telecommunication
IT
0,58
11,53
21,65
4,02
1,84
905,44
75,35 305,45
2,21
1,48
133,61
52,11
20,35
6,67
7,85
398,97
6,98
6,33
5,48
0,51
IT
4,84
39,95
13,47
6,38
1,55
70,14
41,78
17,14
6,60
2,29
Health Care
Health Care
775,34
30,84
10,27
1,19
2,82
207,62
1,33
7,02
2,42
1,83
29296,64 4592,58 597,73
12,31
2,02
6712,65 7591,98 250,82
5,96
4,23
Health Care
Commodities
2900,46
37,81
12,42
10,42
2,10
4396,55
9,46
12,79
10,31
7,73
58,56
79,68
19,38
1,15
1,20
4,97
45,29
29,82
4,70
3,30
Commodities
1233,23
131,78
0,19
8,09
4,54
204,67
8,12
19,89
13,22
2,46
Energy
62
33
63
64
65
List of Figures
Figure 1 - Harry M. Markowitz
3
Figure 2 - Utility Curve for Investors with Different Risk Preferences .................................................. 4
Figure 3 - Combinations of the risk less asset in a risky portfolio (Gruber et al. [9])............................. 9
Figure 4 - The effect of number of securities on risk of the portfolio in the United States (Gruber et al
[9])......................................................................................................................................................... 16
Figure 5 - Relationship between expected return and standard deviation of return for various
correlation coefficients (Gruber et al. [9].)............................................................................................ 17
Figure 6 - The Security Market Line (Gruber et al. [9])........................................................................ 18
Figure 7 - The Capital Market Line (Gruber et al. [9]) ......................................................................... 19
Figure 8 - The Efficient Frontier (Gruber et al. [9]).............................................................................. 21
Figure 9 - Efficient Frontier with RF and M (Gruber et al. [9])............................................................. 22
Figure 10 - PDFs with the same expectation and variance.................................................................... 23
Figure 11 - Different form of Kurtosis. ................................................................................................. 24
Figure 12 - The Solver Optimizer. ........................................................................................................ 27
Figure 13 - The Efficient Frontier. ........................................................................................................ 28
Figure 14 - Portfolio VaR...................................................................................................................... 28
Figure 15 - Spreadsheet fromula behind VaR calculation..................................................................... 29
Figure 16 - Asset Returns' Distribution - Daily data (year 1997-2002). ............................................... 33
Figure 17 - Data Normally Dirtributed.................................................................................................. 34
Figure 18 - Data has fat tails. ................................................................................................................ 34
Figure 19 - Data has short tails.............................................................................................................. 35
Figure 20 - Data is right skewed............................................................................................................ 35
Figure 21 - The normal probability plot for 10 years daily asset returns. ............................................. 36
Figure 22 - Extreme losses in normal VS. Non-normal asset distributions with equal volatility ......... 37
Figure 23 - Volatility and extreme losses of DJIA daily returns. ..................................................... 38
66
List of Tables
Table 1 - Percentage of the diversifiable risk eliminated ...................................................................... 15
Table 2 - Daily Portfolio ....................................................................................................................... 41
Table 3 - Weekly Portfolio.................................................................................................................... 42
Table 4 - Monthly Portfolio................................................................................................................... 43
Table 5 - Quarterly Portfolio ................................................................................................................. 44
Table 6 - Difference Ratios for Different Time Horizons ..................................................................... 46
Table 7 - Sharpe Ratio........................................................................................................................... 47
Table 8 - Portfolio Risk in Different Time Horizons. ........................................................................... 49
Table 9 - Portfolio Return in Different Time Horizon. ......................................................................... 49
Table 10 - Skewness for stocks and different type of data series.......................................................... 59
Table 11 - Kurtosis for stocks and different type of data series. ........................................................... 60
Table 12 - Beta for individual stocks. ................................................................................................... 61
Table 13 - Jarque Bera Test Results. ............................................................................................. 62
67
List of Abbreviations
CAL
CAPM
CML
DJIA
GMV
MPT
MVaR
PDF
SML
VaR
68